Unsupervised Techniques for Meta-analysis of Cancer Genomic Data

Abstract

Cancer is a set of diseases where abnormal cell growth can be identified. The cancer cells are not responding to the cell division or the normal signaling system of human beings. That is the main difficulty in giving treatments and control the growth of those cells. Same cancer has different molecular structures. Different people with the same cancer can show completely different behavior. Personalized medicine is essential for solving that issue. To prepare personalized medicine the identification of subtypes is really important. When considering cancer subtype identification process, the high dimensionality of genomic data is considered as an obstacle. In the biological domain, high dimensionality refers to the high number of genes when compared to the number of samples available. The state of the art methods for dimensionality reduction sometimes does not accurately address this problem when those applied to some of the biological datasets. The performance and the suitability of them vary with the context where we apply those methods. So those techniques should be evaluated through literature and testing with our datasets. Then those techniques will be compared by analyzing their suitability with the genomic data. In this research, a pipeline has been introduced to reduce dimensionality, clustering, and validating which help in cancer subtype identification tasks. The applicability and suitability of the introduced pipeline are evaluated through a set of internal and external validation methods.